Streaming¶

aiterator() reads big results a chunk at a time:

async for book in Book.objects.aiterator(chunk_size=2000):
    await send(book)

• Memory stays flat. The next chunk is read when the loop has used the previous one, so a million rows are never all in memory.
• Nothing waits for the stream. It reads on a connection of its own, so other queries and writes go on, even while the loop is halfway through.
• A consistent view. It sees the database as it was when it started.
• Prefetching works: with prefetch_related(), each chunk is prefetched, like in Django.
• Inside a sync transaction (like Django's TestCase), the rows are read all at once on the transaction's connection, the only one that sees its writes.
• Hooks see it once, when its first chunk arrives (Hooks).